Listed for Sale: Analyzing Data on Fentanyl, Fentanyl Analogs and Other Novel Synthetic Opioids on One Cryptomarket T

Home , Alphamethylthiofentanyl, Bromadoline

Drug and Alcohol Dependence 213 (2020) 108115

Contents lists available at ScienceDirect

Drug and Alcohol Dependence

journal homepage: www.elsevier.com/locate/drugalcdep

Full length article Listed for sale: Analyzing data on fentanyl, fentanyl analogs and other novel synthetic opioids on one cryptomarket T

Francois R. Lamya,*, Raminta Daniulaityteb, Monica J. Barrattc,d, Usha Lokalae, Amit Shethf, Robert G. Carlsong a Department of Society and Health, Faculty of Social Sciences and Humanities, Mahidol University, Salaya, Thailand b College of Health Solutions, Arizona State University, Phoenix, AZ, United States c Social and Global Studies Centre, RMIT University, Melbourne, VIC, Australia d National Drug and Alcohol Research Centre, UNSW Sydney, NSW, Australia e Kno.e.sis Center, Department of Computer Science, Wright State University, Dayton, OH, United States f Artiﬁcial Intelligence Center, University of South Carolina, Columbia, SC, United States g Center for Interventions, Treatment, and Addictions Research, Department of Population and Public Health Sciences, Wright State University, Dayton, OH, United States

ARTICLE INFO ABSTRACT

Keywords: Background: The United States is facing a "triple wave" epidemic fueled by novel synthetic opioids. Cryptomarkets Cryptomarkets, anonymous marketplaces located on the deep web, play an increasingly important role in the Darknet markets distribution of illicit substances. This article presents the data collected and processed by the eDarkTrends Fentanyl platform concerning the availability trends of novel synthetic opioids listed on one cryptomarket. Fentanyl analogs Methods: Listings from the DreamMarket cryptomarket "Opioids" and "Research Chemicals" sections were col- Synthetic opioids lected between March 2018 and January 2019. Collected data were processed using eDarkTrends Named Entity Recognition algorithm to identify opioid drugs, and to analyze their availability trends in terms of frequency of listings, available average weights, average prices, and geographic indicators of shipment origin and destination information. Results: 95,011 opioid-related listings were collected through 26 crawling sessions. 33 novel synthetic opioids were identiﬁed in 3.3 % of the collected listings. 44.7 % of these listings advertised fentanyl (pharmaceutical and non-pharmaceutical) or fentanyl analogs for an average of 2.8 kgs per crawl. "Synthetic heroin" accounted for 33.2 % of novel synthetic opioid listings for an average 1.1 kgs per crawl with 97.7 % of listings advertised as shipped from Canada. Other novel synthetic opioids (e.g., U-47,700, AP-237) represented 22 % of these listings for an average of 6.1 kgs per crawl with 97.2 % of listings advertised as shipped from China. Conclusions: Our data indicate consistent availability of a wide variety of novel synthetic opioids both in retail and wholesale-level amounts. Identiﬁcation of new substances highlights the value of cryptomarket data for early warning systems of emerging substance use trends.

1. Introduction a steady increase in the number of pharmaceutical opioid-related overdose deaths (Paulozzi et al., 2006; Hall et al., 2008) as well as an The United States is experiencing the worst opioid overdose epi- increase in the prevalence of opioid use disorder (Compton and Volkow, demic in its history. The ﬁrst wave of the opioid crisis is associated with 2006). The second wave of the opioid crisis was marked by the in- prescription opioids misuse. It was fueled by the increased volume of creases in heroin use, as growing numbers of individuals who abused opioid prescribing, including by general practitioners, that could be pharmaceutical opioids started transitioning to heroin, a more potent, traced back to the recognition of pain as the "ﬁfth vital sign," increased plentiful and less costly alternative to illicit pain pills (Carlson et al., concerns about the under-treatment of pain, and more aggressive 2016; Cicero et al., 2014; Mars et al., 2014; Unick et al., 2014, 2013). marketing and distribution strategies employed by pharmaceutical Emerging data suggest that some of the measures that were designed to companies (Van Zee, 2009; Ciccarone, 2019). From the mid 1990s to curb the availability of pharmaceutical opioids, might have inad- early 2010s, the diversion and misuse of pharmaceutical opioids caused vertently accelerated transition to heroin use among individuals who

⁎ Corresponding author. E-mail address: [email protected] (F.R. Lamy). https://doi.org/10.1016/j.drugalcdep.2020.108115 Received 12 February 2020; Received in revised form 14 May 2020; Accepted 8 June 2020 Available online 12 June 2020 0376-8716/ © 2020 Elsevier B.V. All rights reserved. F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115 were dependent on pharmaceutical opioids (Castillo-Carniglia et al., beginning of 2019 (Ball et al., 2019). However, information regarding 2019). Finally, since 2013–2014, novel synthetic opioids, such as illi- the origin and type of these novel synthetic opioids are needed to citly manufactured fentanyl, fentanyl analogs (i.e., substances sharing characterize cryptomarket data related to specific types of novel syn- similar chemical structure and action as fentanyl) and other novel thetic opioids, such as specific types of fentanyl analogs and other novel synthetic opioids (e.g., U-47,700, MT-45) have emerged in the drug synthetic opioids. To establish timely and effective policy and inter- market creating what has been called a "third wave" of the opioid vention responses to the opioid crisis in the U.S., epidemiological epidemic in the US (Ciccarone, 2019; Jones et al., 2018) and exacer- monitoring systems need to be able to rapidly identify shifts in avail- bating the opioid crisis even further. ability trends of specific novel synthetic opioid products and alert In the U.S, novel synthetic opioids were responsible for 30.5 % (n = policy makers about emerging new substances. This study builds on the 74,409) of all opioid-related overdose deaths that occurred nationally eDarkTrends project that was developed to collect and process cryp- between 2010 and 2017 (Hedegaard et al., 2018). The negative impacts tomarket data on fentanyl, fentanyl analogs and other novel synthetic in some of the Northeastern and Midwestern communities were even opioids. The overall aims of this paper are to present data collected more profound (Pardo et al., 2019). For example, in Montgomery from Dream Market over 10 months, focusing on the amounts of drugs County, Ohio, about 90 % of all unintentional overdose deaths that offered for sale (in terms of number of listings and total weight avail- occurred in 2016 and 2017 tested positive for novel synthetic opioids able for purchase), prices, and information on distribution patterns in (Daniulaityte et al., 2017a,b, 2019). terms of the countries listed as shipment origin and destination. Novel synthetic opioids are a part of the larger group of new psychoactive substances defined by the United Nations Office on Drugs and 2. Methods Crime (UNODC) as "substances of abuse, either in a pure form or a preparation, that are not controlled by the 1961 Single Convention on 2.1. Data collection Narcotic Drugs or the 1971 Convention on Psychotropic Substances, but which may pose a public health threat." Between 2010 and 2019, the eDarkTrends, a semi-automated platform, was developed to monitor European Monitoring Centre for Drug and Drug Addiction (EMCDDA) listings of fentanyl, fentanyl analogs, and other novel synthetic opioids identified 49 novel synthetic opioids in the European drug markets on cryptomarkets to assess availability trends and identify emerging (EMCDDA, 2019). Most of these substances were not "new" per se, but substances. The data presented in this paper were collected from the rather, a resurgence of substances from older biomedical literature with DreamMarket cryptomarket and covered a nearly one-year time period limited information on their side effects (Baumann et al., 2018). Al- from March 2018 to January 2019, until DreamMarket was shut down though fentanyl was part of the substances controlled under the 1961 in February 2019 (Madore, 2019). At that time, DreamMarket was the convention, and has a long history of medical use in a variety of largest existing cryptomarket with over 60,000 listings for drugs pharmaceutical forms (e.g., Duragesic transdermal system), it is in- (Knight, 2019). Although it established a ban on fentanyl and fentanyl cluded under the label of novel synthetic opioid because current in- analog listings in May of 2018 (Ball et al., 2019), data collection was creases do not involve pharmaceutical fentanyl, but are associated with maintained as the DreamMarket remained the biggest cryptomarket non-pharmaceutical fentanyl that is illicitly produced in clandestine advertising thousands of other illicit synthetic opioids, and some ven- laboratories. It also reflects the actual position of the Drug Enforcement dors continued to sell fentanyl-type drugs under other names. Agency (DEA), which considers fentanyl as the "most prevalent and Data were collected using a custom eDarkTrends web crawler de- most significant synthetic opioid threat in the United States" (DEA, veloped and maintained by the research team at the Kno.e.sis Center 2018a,b). (fourth author U.L.). A crawler is an automated script that system- There is a growing body of evidence showing that darknet markets, atically collects all the information contained on the selected websites also known as cryptomarkets, play an increasingly important role in the or their sub-sections. The eDarkTrends web crawler targeted the distribution of illicit substances and goods in Western countries "Opioids" and "Research Chemicals (RC's)" sections of DreamMarket and (Dolliver and Kuhns, 2016; Kruithof et al., 2016a, b), despite recent law excluded other sections. Crawler’s outputs were raw HyperText Markup enforcement operations that have dismantled several darknet markets Language files (HTML) of drug advertisements with image source at- (Madore, 2019). In contrast to the "surface web," the "deep web" is a tributes removed. These raw HTML files contained the following in- section of the Internet where websites or databases are not listed by formation: product listing, vendor pseudonym, vendor number of sales search engines such as Google, Bing or Yahoo. It encompasses all pri- and level of trust, information about the product provided by the vate data accessible via the Internet and represents approximately 96 % vendor, price (Bitcoin and US$), country/region of shipment origin, of the total web. The "darknet" is a small subsection of the deep web potential shipment destination country/region, security measures con- that can only be accessed by using an anonymizing browser (e.g., The cerning transactions, terms and conditions, and reviews from previous Onion Router (Tor) or Invisible Internet Project (I2P)) providing en- buyers. Data extraction, storage and access procedures followed strict crypted communication and a decentralized network structure security measures reviewed and approved by the Wright State (Aldridge and Décary-Hétu, 2016a,b; Barratt et al., 2018). Crypto- University’s Information Security Office. markets are online marketplaces that function within the darknet, using Data were collected at 26 time points. Although initial plans in- cryptocurrencies (e.g., Bitcoin, Monero, Litecoin) and advanced en- cluded data collection on a more frequent basis, the eDarkTrends cryption techniques to off er anonymity to vendors and consumers platform experienced difficulties in completing more frequent crawls trading for illegal goods or services (Barratt and Aldridge, 2016). Over because of the technical challenges and extreme volatility of crypto- 60 % of the listings on cryptomarket are dedicated to pharmaceutical market environments. For example, DreamMarket was not always ac- and/or illicitly produced psychoactive substances (i.e., pharmaceu- tive due to maintenance or because of DDoS (Distributed Denial of ticals, drug-related chemicals or illicit drugs) (Europol and EMCDDA, Service) attacks that aim to overload servers with requests to disrupt its 2017). normal functioning. Prior studies have monitored cryptomarkets to identify emerging NPS and other psychoactive drugs by evaluating the amount of sub- 2.2. Data processing stances being supplied, drug prices, number of sellers, seller revenue, number of cryptomarkets, and country-specific information (Dolliver, Extracted data were processed to identify relevant drug mentions 2015; Kruithof et al., 2016a,b; Van Buskirk et al., 2016). A recent sta- using the eDarkTrends-dedicated Named Entity Recognition (NER) al- tistical bulletin described the overall amount, price and physical forms gorithm developed by the first author (more details in Lokala et al., of fentanyl available on six cryptomarkets over a 51-day period in the 2019). The NER algorithm was developed in the Python programming

2 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115 language (van Rossum, 1990) and integrated three key components: (1) pharmaceutical forms was classified as non-pharmaceutical fentanyl the Natural Language ToolKit (NLTK) library was used to curate and (e.g., "Fentanyl HCl crystalls pure 98 % 1 g"). The NER was also de- process text segments from crawled data (Bird et al., 2009); (2) the veloped to correctly identify misspellings, abbreviations and slang ter- Drug Abuse Ontology (DAO) that functions as a domain-specific con- minology for fentanyl analogs and other novel synthetic opioids (e.g., ceptual framework for interconnecting sets (named “classes”) of drug- carfentanyl as carfentanil; 3-MMAF as 3-methylmethoxyacetylfentanyl; focused lexicons to provide a list of terms to be identified (Cameron butyrylfentanyl as butyrentanyl). et al., 2013; Daniulaityte et al., 2017a,b; Lamy et al., 2017), and; (3) To minimize classifi cation errors, the following strategies were used Regular Expressions (i.e., a sequence of symbols and characters forming for more ambiguous drug listing scenarios. Listings for oxycodone a pattern to be searched within a text or sentence) built on the selected “pressed pills” containing fentanyl (e.g., "QTY: 50 Oxycodone 30 mg entities populating the DAO to capture objects of interest. A215 FENTANYL PRESSED") were categorized as non-pharmaceutical The Drug Abuse Ontology (DAO) was developed by the research fentanyl under a separate "Oxycodone pills with fentanyl" subcategory. team building on our prior work on web-based analytics for substance Listings offering sale of "China White" with a mention of fentanyl or use epidemiology research (Cameron et al., 2013; Daniulaityte et al., fentanyl analogs (e.g., "This listing is for synthetic heroin also known as 2015; Lamy et al., 2017). The DAO was created using Protégé software china white. This is a mix of Furanyl-Fentanyl and Pharma grade (Musen, 2015) and includes a comprehensive list of psychoactive sub- mannitol professionally mixed in a mixer") were classified as non- stances based on available resources (e.g., Drug Enforcement Admin- pharmaceutical fentanyl or fentanyl analog posts. Although slang term istration, National Institute on Drug Abuse, European Monitoring "China White" is sometime used to refer to fentanyl that is sold as heroin Centre for Drugs and Drug Addiction), prior research (e.g., Daniulaityte (Martin et al., 1991; Ciccarone et al., 2017), but it can also refer to et al., 2015; Lamy et al., 2017), and preliminary examination of cryp- heroin products (Ciccarone et al., 2017). China White listings without tomarket data. The DAO includes names of psychoactive substances any mention of another novel synthetic opioids were not classified as (e.g., heroin, fentanyl, acetylfentanyl) including synthetic substances belonging to any of the novel synthetic opioid categories. "Synthetic (e.g., U-48,800, MT-45), brand and generic denominations of pharma- heroin" listings lacked sufficient information to classify them together ceutical drugs (e.g., Fentora, Subutex) and slang terms (e.g., china with non-pharmaceutical fentanyl, fentanyl analogs or other novel white, oxy, fent). It also contains information regarding the route of synthetic opioids, and thus they are presented as a separate, more administration (e.g., transdermal, IV), unit of dosage (e.g., gr, kg, ml, ambiguous category. However, there is evidence to suggest that the films), physical forms (e.g., powder, patch, drop), and drug effects (e.g., term "synthetic heroin" is used to refer to fentanyl and/or other fentanyl euphoria, pain control). analogs mixed with or sold as heroin in lay and popular media termi- The eDarkTrends NER algorithm was designed to identify and group nology (Ciccarone et al., 2017; Rosenblum et al., 2020). opioid drug mentions into higher-level classes (e.g., heroin, pharma- The eDarkTrends NER also identified the amount of substances ceutical fentanyl, non-pharmaceutical fentanyl). These classes included listed for sale in terms of weight (i.e., total amount in gram or milliliter) sub-classes to identify specific types of fentanyl analogs and other and quantity (i.e., total number of items). Weight and units of dosage drugs. For example, the fentanyl analog class included a list of 51 were transformed into grams (g) for powder and base form, in milliliter fentanyl analog and precursor names (e.g., carfentanyl, acetylfentanyl), (ml) for liquid substances. For pharmaceutical fentanyl or counterfeit and other novel synthetic opioid class included a list of 17 compounds pharmaceutical products (e.g., films, patches, pills) as well as for oxy- (e.g., U-47,700, AP-237) (see Table 1). Entities that were not re- codone pills pressed with fentanyl, the amount was calculated based on cognized by the NER algorithm were grouped into a separate “Uni- the number of items (quantity) rather than the weight to ensure that dentified” class. listings with a large number of items that contain a very small amount For the fentanyl-type drug class, the eDarkTrends NER algorithm of active ingredient (e.g., patches with micrograms of fentanyl) would identified both pharmaceutical and non-pharmaceutical fentanyl. not distort average price calculations. Duplicate listings were not re- Listings that mentioned pharmaceutical fentanyl brand names (e.g., moved from the analysis for two main reasons: a) DreamMarket listings Duragesic, Actiq, Sublimaze) or standard pharmaceutical forms of ad- did not offer the possibility for vendors to display their stock or ministration (e.g., transdermal system, lollipop, spray) were classified "quantity left" such as in some other markets (e.g., EmpireMarket); 2) as pharmaceutical fentanyl (e.g., "Subsys Sublingual Fentanyl Spray new listings were added or removed over time suggesting that listings 200mcg Pharmacy"). Fentanyl in powder form or other non- were regularly updated and did not accumulate over time such as in

Table 1 Categorization of opioid listings.

Category Sub-category

Heroin Heroin “Synthetic Heroin” Synthetic Heroin Pharmaceutical Fentanyl Pharmaceutical Fentanyl (e.g., Duragesic, Sublimaze, fentanyl transdermal system) Non-Pharmaceutical Fentanyl 1)Non-pharmaceutical fentanyl; 2) Oxycodone pills with fentanyl Fentanyl Analogs 1) 3-allylfentanyl; 2) 3-methylbutyrfentanyl; 3) 3-methylfentanyl; 4) 3-methylmethoxyacetylfentanyl; 5) 3-methylthiofentanyl; 6) 3m- furanylfentanyl, 7) 3m-methoxylacetylfentanyl; 8) 4anpp; 9) 4appp; 10) 4-fluoroisobutyrfentanyl; 11) 4-flurobutyrfentanyl; 12) 4- methoxymethylfentanyl; 13) 4npp; 14) 4-phenylfentanyl; 15) acetylfentanyl; 16) acrylfentanyl; 17) acryloylfentanyl; 18) alfentanyl; 19) alphamethylthiofentanyl; 20) amethylacetylfentanyl; 21) appp; 22) benzylfentanyl; 23) betahydroxyfentanyl; 24) betahydroxythiofentanyl; 25) betamethylfentanyl; 25) butryfentanyl; 27) butyrfentanyl; 28) butyrfluorofentanyl; 29) butyrfuranylfentanyl; 30) carfentanil; 31) crotonylfentanyl; 32) etorphine; 33) etorphinecartanil; 34) fluorofentanyl; 35) furanylfentanyl; 36) isobutyrfentanyl; 37) lofentanyl; 38) methoxyacetylfentanyl; 39) methylfentanyl; 40) mirfentanyl; 41) nmethylcarfentanil; 42) ocfentanil; 43) ohmefentanyl; 44) orthomethylfuranylfentanyl; 45) paraflurofentanyl; 46) pentafentanyl; 47) R-30,490; 48) remifentanil; 49) sufentanil; 50) tetrahydrofuranylfentanyl; 51) thiofentanyl Other novel synthetic opioids 1) U-50,488; 2) U-47,700; 3) U-49,900; 4) U-48,800; 5) MT-45; 6) AH-7921; 7) W-18; 8) MPF-47700; 9) U-4TDP; 10) isopropyl-U-47,700; 11) MP-47700; 12) U-47v700; 13) U-47,931e; 14) FS-33; 15) U-57,700; 16) AP-237; 17) W-15 Other Pharmaceutical opioids 1) buprenorphine; 2) codeine; 3) hydrocodone; 4) hydromorphone; 5) loperamide; 6) methadone; 7) morphine; 8) naloxone; 9) oxycodone; 10) oxymorphone; 11) tramadol Kratom Kratom Opium Opium

3 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115

Table 2 Total numbers of heroin, fentanyl, fentanyl analog and other novel synthetic opioid listings, main shipment origins, percentage shipped to the United States and associated unique vendor counts on DreamMarket cryptomarket during the eDarkTrends data collection period (22nd of March 2018 to 26th of January 2019).

Substances Number of Listings Percent of all Main countries of origin (% of % shipping to the Number of Unique (TOTAL N = 95,011) listings listings) U.S.A. Vendors

Heroin 52,510 55.27% U.S.A., 27.0 %; Germany, 22.1 %; 80.00 % 532 U.K., 13.8 % “Synthetic Heroin” 1031 1.09% Canada, 97.9 %; U.S.A., 2.1 % 100 % 30 Pharmaceutical Fentanyl* 301 0.32% Italy, 32.4 %; Netherlands, 20.7 %; 59.90 % 36 Germany, 19.6 % Non-Pharmaceutical Fentanyl Non-Pharmaceutical Fentanyl 395 0.42% Australia, 38.3 %; U.S.A., 29.5 %; 93.40 % 33 Canada, 21.8 % Oxycodone Pills with Fentanyl* 123 0.13% U.S.A., 100 % 100 % 5 Fentanyl Analogs Furanylfentanyl (Fu-F) 217 0.23% Slovakia, 66.1 %; Germany, 21.0 %; 94.00 % 5 India, 1.9 % Methoxyacetyl-fentanyl (MAF) 103 0.11% China, 39.1 %; India, 32.6 %; U.S.A., 93.20 % 12 26.1 % 4-Fluoroisobutyr-fentanyl (4-FIBF) 101 0.11% China, 63.6 %; India, 36.4 % 100 % 4 Carfentanil 45 0.05% China, 52.5 %; U.S.A., 40.0 %; 100 % 7 Germany, 7.5 % Butyrﬂuorofentanyl 35 0.04% China, 100 % 85.70 % 2 Precursors (4-ANPP, 4-APPP, APPP) 20 0.02% China, 100 % 100 % 1 Acetylfentanyl 13 0.01 % U.K., 100 % 100 % 1 4-Phenylfentanyl 12 0.01 % ND 100 % 1 3-Methylmethoxy-acetylfentanyl (3- 10 0.01 % China, 100 % 100 % 1 MMAF) Butyrfentanyl (Butyrylfentanyl) 6 0.01 % ND 33.30 % 2 4-Methoxymethyl-fentanyl (R-30490) 4 0.00% China, 66 %; India, 33 % 100 % 2 Orthomethyl-furanylfentanyl (o-Me-Fu-F) 2 0.00% U.S.A., 100 % 100 % 1 Other Novel Synthetic Opioids U-48,800 224 0.24% China, 64.3 %; Australia, 3.1 % 93.70 % 13 U-50,488 106 0.11% China, 100 % 100 % 3 U-4TDP 100 0.11% China, 100 % 100 % 1 U-47,700 84 0.09% China, 79.8 %; India, 3.6 %; U.S.A., 100 % 6 2.4 % Isopropyl-U-47,700 59 0.06% China, 8.5 % 100 % 1 MPF-47,700 42 0.04% China, 100 % 100 % 1 MP-47,700 37 0.04% China, 100 % 100 % 1 U-49,900 9 0.01 % China, 66.6 % 66.60 % 2 U-58,800 6 0.01 % China, 100 % 100 % 1 U-47v700 4 < 0.01 % China, 100 % 100 % 1 AP-237 3 < 0.01 % China, 100 % 100 % 2 Bromadoline (U-47931e) 3 < 0.01 % China, 100 % 100 % 1 XIX-18 2 < 0.01 % China, 100 % 100 % 1 Etorphinecartanil 1 < 0.01 % China, 100 % 100 % 1 FS-33 1 < 0.01 % ND 100 % 1 U-45,500 1 < 0.01 % ND 100 % 1 W-18 1 < 0.01 % China, 100 % 100 % 1

Alphabay. Instead, we have assumed that vendors posting similar list- selected for manual assessment of the accuracy of the eDarkTrends ings were advertising a specific weight/quantity of a product they had Named Entity Recognition (NER) algorithm. The first author manually in stock several times. classified each of the selected listings (n = 555) to identify the class and Prices were converted from Bitcoin (BTC) to U.S. dollars according sub-class of the listed substance, its weight, form and quantity. The to the BTC-$US conversion rate on the day DreamMarket was crawled. results of the manual classification were compared to the results of Listings offering free drug samples were excluded when calculating the automated classification performed by the eDarkTrends NER algorithm. average price. Average prices (U.S. $ per gram) were calculated sepa- The accuracy assessment indicates that the NER algorithm was able to rately for retail and wholesale-level amounts. The weight limit between correctly identified 99.3 % of substances, 98.9 % of class of substances, "retail" and "wholesale" quantities was arbitrarily set to 5 g (20 items for 96.9 % of weight, 98.2 % of form and 91.0 % of quantity contained in listing not in powder or liquid form) to acknowledge the potency of the listings from the random subsample. novel synthetic opioids. Shipment origins and destinations were grouped by country. Manual checks have been performed to remove 3. Results inconsistent data and outliers to prevent potential distortions in the calculations of averages. Listings displaying several countries (e.g., 3.1. Collected data "FRANCE BELGIUM HOLLAND GERMANY") or imprecise/inexistent/ ambivalent countries of origins (e.g., "North Pole," "The White House") 110,892 listings were collected from DreamMarket through 26 were excluded from the analysis. crawling sessions between the 22nd of March 2018 and 26st of January 2019. 85.7 % (95,011/110,892) of all collected listings were identified 2.3. Reliability assessment of the named entity recognition algorithm as related to opioids, and the remaining 14.3 % (15,881/110,892) did not contain elements that the NER identified as belonging to the opioid A random subsample of 555 listings (0.5 % of all listings) was class of substances (i.e., non-opioid drugs, special customer order or "tip

4 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115 jar" for vendor). These listings were excluded from further analysis. On opioids, namely, 4-methoxy-methylfentanyl, butyrfluorofentanyl, 4- average, 4,107 listings related to opioids were collected at each crawl. APPP, APPP, 4-phenylfentanyl, orthomethylfuranyfentanyl (o-ME-Fu- F), U-50,488, U-4TDP, MPF-47,700, MP-47,700, U-58,800, U-47v700, 3.2. Availability of novel synthetic opioids bromadoline (U-47931e), XIX-18, etorphinecartanil, FS-33, U-45,500 and W-18, haven't been identified in the seizures/evidences tests run by Out of 95,011 opioid-related listings, 3.3 % (3,101/95,011) were for the DEA nor in the forensic tests run by the NMS Labs at the time this novel synthetic opioids. In comparison, heroin was identified in 55.2 % article was written. AP-237 was identified by NMS labs in September (52,510/95,011) of all opioid-related listings, and there were 1,945 2019 (NMS Labs, 2019b). heroin-related listings per each crawl, on average. Both pharmaceutical fentanyl (301/95,011) and non-pharmaceutical fentanyl and analogs (1,086/95,011) were advertised on 3.3. Prices DreamMarket representing 1.5 % (1,387/95,011) of all opioid-related listings and accounting for 44.7 % (1,387/3,101) of all novel synthetic The average price of a gram of non-pharmaceutical fentanyl was opioid-related listings. During the study period, an average of 2.8 kg of U.S.$1,470.4 for listings offering less than 5 g and U.S.$139.5 for list- fentanyl-type substances were listed for sale at each time point of data ings offering 5 g or more. Notably, carfentanil was advertised for an collection. In addition, there were an average of 2,495 non-pharma- average price of U.S.$2,311.8 per gram for retail-level amounts and an ceutical oxycodone pills pressed with fentanyl available for purchase at average price of U.S.$101.0 per gram for wholesale-level amounts. each time point when data were collected. The information about the Furanylfentanyl (Fu-F), the most common fentanyl analog appearing on amount of fentanyl added to these pills was specified only by one DreamMarket, was advertised for an average price of U.S.$243.1 per vendor, in a total of 7 listings (out of 123), and was equal to 150 μg. gram for retail listings and an average price of U.S.$26.5 per gram for "Synthetic heroin" was the most frequent type of other novel syn- wholesale listings. For other novel synthetic opioids, U-48,800 was thetic opioid advertised accounting for 1.1 % (1,031/95,011) of all advertised on average U.S.$48.4 per gram for retail-level amounts and opioid-related listings, and representing 33.2 % (1,031/3,101) of all U.S.$14.2 per gram for listings offering wholesale-level amounts. U- novel synthetic opioid-related listings. On average, there was about 1.1 47,700 was offered for an average price of U.S.$100.3 per gram for kg of synthetic heroin listed for sale at each data point. Remaining other retail listings and an average price of U.S.$19.7 per gram for wholesale novel synthetic opioids products (e.g., W-18, U-47,700) were identified listings. Finally, "synthetic heroin" was advertised for an average price in 0.7 % (683/95,011) of all opioid-related listings. They comprised of U.S.$121.2 per gram for retail listings and an average price of U.S. about 22 % (683/3,101) of all novel synthetic opioid-related listings, $69.5 per gram for wholesale listings. For comparison, the average with an average of 6.1 kg offered for sale at each data point (Table 2). price for a gram of heroin was U.S.$206.5 for retail-level and U.S.$32.3 Eleven fentanyl analogs (acetylfentanyl, burtyrfentanyl, butyr- for wholesale-level amounts (Table 3). fluorofentanyl, carfentanil, furanylfentanyl, methoxyacetylfentanyl, orthomethylfuranylfentanyl, 3-methylmethoxyacetyl-fentanyl, 4-fluor- obutyrfentanyl, 4-methoxymethylfentanyl (R-30,490) and 4-phe- 3.4. Shipment origins and destinations nylfentanyl) and three precursor chemicals (advertised as 4-ANPP, 4- APPP and APPP) were identified over the data collection period. Among More than half (53.7 %, 51,018/95,011) of all opioid-related list- fentanyl analogs, furanylfentanyl was the most commonly advertised ings displayed country-level geolocation information indicating ship- (217/568) with an average of 2.3 kg offered for sale at each data point, ment origin ("shipping from"). Out of all fentanyl-related listings (both followed by methoxyacetyl-fentanyl (103/568), and 4-fluor- pharmaceutical and non-pharmaceutical), 66.9 % (928/1,387) dis- oisobutyrfentanyl (101/568). There was a total of 45 listings (out of a played country-level geolocation information. Notably, the United total of 568 fentanyl analog listings) for carfentanil, an extremely po- States was the main country of origin from where vendors indicated tent fentanyl analog that is also used in veterinary medicine as a large they were selling non-pharmaceutical oxycodone pills mixed with fen- animal tranquilizer. On average, there were 548.7 g of carfentanil tanyl with 100 % (116/116) of the listings displaying geolocation in- available for purchase during the data collection period (Table 2). formation at the country level (116 out of the total 123 listings). The fentanyl ban implemented by DreamMarket on the 20th of May Furthermore, non-pharmaceutical fentanyl was mainly advertised by 2018 had considerable effect on the availability of non-pharmaceutical unique vendor names indicating they were shipping from Australia fentanyl and fentanyl analog listings. Before the ban, there was an (38.3 %, 114/298), followed by the United States (29.5 %, 88/298), average of 144.8 listings (and average weight of 11.1 kg) for non- and Canada (21.8 %, 65/298). pharmaceutical fentanyl or fentanyl analogs at each time period of data Most of the fentanyl analogs were predominantly advertised as collection. After the ban, the average number of listings of these sub- shipped from China, except acetylfentanyl (U.K, 100 %, 13/13), or- stances at each data collection period declined to 8.9, and the average thomethylfuranylfentanyl (U.S., 100 %, 2/2) and furanylfentanyl total available weight decreased to 233.8 g. (Slovakia, 66.1 %, 41/62; Germany, 21.0 %, 13/62). China was also the Regarding other novel synthetic opioids, 17 different substances main country of origin for carfentanil with 52.5 % (21/40) of listings were identified in a total of 683 listings. U-48,800 was the most fre- with geolocation at the country level, followed by the U.S. (40.0 %, 16/ quent substance listed among them (32.8 %, 224/683) with an average 40) and Germany (7.5 %, 3/40). Furthermore, China (97.2 %) was the of 1,845.4 g available for purchase at each date of data collection. Eight main shipment origin for other novel synthetic opioids advertised on other novel synthetic opioids (U-47v700, AP-237, bromadoline (U- DreamMarket. Other novel synthetic opioid listings were posted by 24 47,931e), XIX-18, etorphinecartanil, FS-33, U-45,500 and W-18) were vendors with 20 out 24 of them based in China. Only 8 vendors posted identified at only one data point. 10 listings or more, with three of them posting 82.9 % of all other novel Overall, the eDarkTrends NER identified 14 substances (namely, 3- synthetic opioid-related posts. Finally, the majority of the "synthetic methylmethoxyacetylfentanyl (3-MMAF), 4-ANPP, 4-FIBF, acet- heroin" listings with country-level geolocation information were ad- ylfentanyl, butyrylfentanyl, carfentanil, despropionyl-p-fluorofentanyl, vertised as shipped from Canada (97.9 %, 939/959). fentanyl, furanylfentanyl, isopropyl-U-47,700, methoxyacetylfentanyl, More than 80 % of listings advertising novel synthetic opioids on U-47,700, U-48,800, U-49,900) of the 34 novel synthetic opioids listed DreamMarket (with the exceptions of pharmaceutical fentanyl, bu- by the Special Testing and Research Laboratory of the Drug tyrfentanyl, and U-49,900) could be shipped to the United States Enforcement Agency (DEA, 2018a,2018b, 2019) and the National (Table 2). Medical Services (NMS Labs, 2019a). The remaining 16 novel synthetic

5 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115

Table 3 Average total available weights, numbers or wholesale and retail listings and average prices for heroin, fentanyl, fentanyl analogs and other novel synthetic opioids advertised on DreamMarket during the eDarkTrends data collection period (22nd of March 2018 to 26th of January 2019). * Diﬀerence between Retail and Wholesale was set to 20 items.

Substances Average total weights available at each data Retail Amounts (< = 5 g) Wholesale Amounts (> 5 g) collection time point – grams (min-max) Average price in US$ per Number of Average price in US$ per Number of gram (min-max) listings gram (min-max) listings

Heroin 954.8 (601−2,267) 206.5 (28.3−1,881) 24,758 32.3 (0.2−170.2) 27,752 “Synthetic Heroin” 1,149.3 (25.9−4,216.8) 121.2 (0.1−701.9) 643 69.5 (7.0−129.9) 388 Pharmaceutical Fentanyl* 117.2 (5−277) 91.1 (17.2−307.5) 257 75.7 (30.0−109.5) 44 Non-Pharmaceutical Fentanyl Non-Pharmaceutical Fentanyl 266.4 (0.3−1,998.8) 1470.4 (216.7−4,371.9) 274 139.5 (4.3−1,637.3) 57 Oxycodone Pills with Fentanyl* 2,494.7 (25−5,250) 6.4 (1.0−11.9) 6 1.1 (1.0−12.8) 117 Fentanyl Analogs Furanylfentanyl (Fu-F) 2,296.6 (1−6,716.3) 243.1 (13.5−1,442.5) 111 26.5 (4.5−151.6) 106 Methoxyacetyl-fentanyl (MAF) 722.3 (< 0.01−2,594.2) 194.1 (28.0−606.3) 26 40.0 (5.2−165.8) 77 4-Fluoroisobutyr-fentanyl (4-FIBF) 3,265.7 (540.7−3,955.7) 181.8 (17.7−516.5) 47 18.3 (6.6−182.5) 54 Carfentanil 548.7 (0.05−3,551.5) 2,311.8 (182.5−4,221.9) 25 101.0 (26.0−231.6) 20 Butyrﬂuorofentanyl 1977 (1,935−2,040) 33.3 (32.2−34.3) 2 12.3 (6.1−28.9) 33 Precursors (4-ANPP, 4-APPP, 643.7 (100−900) ND 0 5.4 (3.5−7.6) 20 APPP) Acetylfentanyl 1.7 (0−1.7) 134.1 (108.6−159.7) 1 ND 0 4-Phenylfentanyl 1 (0.5−1.5) 1,746.8 (1,694.2−1,799) 1 ND 0 3-Methylmethoxy-acetylfentanyl 510 (510−510) ND 0 17.8 (3.9−34.9) 10 (3-MMAF) Butyrfentanyl (Butyrylfentanyl) 110.1 (110−110.3) 651.2 (579.0−723.4) 2 23.0 (20.5−25.5) 4 4-Methoxymethyl-fentanyl (R- 195 (195−195) ND 0 22.9 (20.8−25.1) 4 30490) Orthomethyl-furanylfentanyl (o- < 0.01 95.2 (75.2−115.2) 2 ND 0 Me-Fu-F) Other Novel Synthetic Opioids U-48,800 1,845.4 (10−3,715) 48.4 (27.5−193) 35 14.2 (4.7−122.3) 189 U-50,488 1,447 (10−1,910) 35.0 (19.2−52.2) 4 10.8 (6.1−32.1) 102 U-4TDP 1,460 (260−1,910) ND 0 10.6 (6.1−22.7) 100 U-47,700 540 (0.06−1,947) 100.3 (43.9−149.0) 27 19.7 (5.9−53.0) 57 Isopropyl-U-47,700 1,425 (885−1,941) 86.4 (43.1−151.9) 12 16.2 (6.6−36.0) 47 MPF-47,700 760 (10−1,912) 25.7 (24.6−26.7) 5 13.1 (6.2−24.3) 37 MP-47,700 1,268 (652-1,902) 36.5 (34.7-38.2) 6 8.6 (6.2-12.8) 31 U-49,900 22.5 (10-110) ND 0 29.5 (23.2-45.8) 9 U-58,800 1,125 (1,125−1,125) ND 0 8.2 (6.2−11.7) 6 U-47v700 1,160 (1,160−1,160) ND 0 17.1 (7.3−31.2) 4 AP-237 10 (10−10) ND 0 41.8 3 Bromadoline (U-47931e) 112.5 (100−125) ND 0 15.3 (7.2−16.6) 3 XIX-18 35 (35−35) 129.9 1 96.6 1 Etorphinecartanil 1 1,989.50 1 ND 0 FS-33 1 ND ND ND ND U-45,500 500 ND 0 9.1 1 W-18 25 ND 0 22.9 1

3.5. Shipment origins by wholesale and retail level amounts 4. Discussion

In terms of the weights advertised for non-pharmaceutical fentanyl Our study presents a detailed description of the type, average and analogs (fentanyl precursors excluded), 34.0 % (363/1,066) of amount and shipment origin of novel synthetic opioids advertised on listings were advertising wholesale quantities (> 5 g). 52.3 % (190/ one cryptomarket (DreamMarket) over a 10-months period (March 363) of them displayed geolocation information at the country level: 2018-January 2019). It is one of the ﬁrst studies to describe and 47.4 % (90/190) advertised as shipped from China, 14.7 % (28/190) quantify the types, amounts, available weights, prices and shipment from India, 12.6 % (24/190) from Australia, and 10.5 % (20/190) from origins and destination of fentanyl, fentanyl analogs and other novel the United States. Among these 363 “wholesale” listings, 150 advertised synthetic opioid products sold on the darknet. Fentanyl, fentanyl ana- quantities equal to or greater than 100 g with 53.3 % (80/150) of them logs, "synthetic heroin" and other novel synthetic opioid substances displaying geolocation at the country-level. China was indicated as the represent only 3.3 % of all opioid listings advertised on DreamMarket shipment origins for 67.5 % (54/80) of these listings, followed by India from March 2018 to January 2019. Although novel synthetic opioids with 20.0 % (16/80), Slovakia with 7.5 % (6/80) and the United States constitute a small fraction of the overall illicit opioid market on the with 1.3 % (1/80). Among the 569 non-pharmaceutical fentanyl and darknet, the potency of these opioids make the amount needed for one analogs (fentanyl precursors excluded) advertising less than 5 g of dose smaller compared to other opioids (Pardo et al., 2019). It also substance, 59.4 % (338/569) displayed geolocation at the country decreases the quantity needed to be sent by mail, lowering the chance level. 39.9 % (135/338) of these listings advertised shipment from the for shipment to be detected and seized by customs or law enforcement U.S, 24.3 % (82/338) from Australia, and 18.3 % (62/338) from agents (Baum, 2017; O’Connor, 2017). Canada. Only 2.1 % (7/338) were advertised as shipped from China. Furthermore, our data suggest that as listed on DreamMarket, the average wholesale price per gram of fentanyl analogs (except carfentanil and methoxyacetylfentanyl) and U-47,700 analogs (e.g., U- 48,800, MPF-47,700, U-58,800) were lower compared to the average

6 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115 wholesale prices for a gram of heroin, making these novel substances a partial and, in some cases, may not represent the real origin of ship- more potent and profitable alternative to heroin for drug dealers. ment, which, in turn, could have contributed to bias in the shipment Interestingly, most of the U-47,700 analogs, such as U-48,800, were origin calculation. Finally, the weight used to differentiate retail from generally sold for lower prices than fentanyl-type drugs, which could wholesale quantities was set arbitrarily and requires further investiga- be, at least, partially explained by the lower potency of these sub- tions to allow a more precise quantification. stances, compared to fentanyl-type drugs: for example, U-47,700 is considered to be 7.5 stronger than morphine, compared to fentanyl considered to be 50–100 times more potent than morphine (Mohr et al., 5. Conclusions 2016). Moreover, the large discount existing between retail and wholesale quantities (up to 23 times for carfentanil and 17 times for Our study provides new insights into the role of the darknet in the non-pharmaceutical fentanyl) indicates substantial incentives for prof- availability of non-pharmaceutical fentanyl, fentanyl analogs and other itability of retail-level activities and distribution. novel synthetic opioids. Although focusing on one market DreamMarket advertised a large amount and variety of opioids with (DreamMarket) over a 10-months period, data indicate a persistent more than 4000 opioid-related posts advertising as much as 71 different availability of both retail and wholesale quantities of non-pharmaceu- types of opioid products on a regular basis. During the data collection tical fentanyl, fentanyl analogs, and other synthetic opioids, drugs that period, 33 substances classified as novel synthetic opioids were iden- are causing devastating increases in unintentional drug overdose deaths tified by the eDarkTrends platform, and at the time of writing this in the U.S. (Peterson et al., 2016; Scholl et al., 2019) and in Canada paper, 17 of them have not yet been identified by the DEA or NMS in (Hung et al., 2019). The exact role of the darknet in novel synthetic the United States. However, novel synthetic opioid AP-237 was first opioid availability continues to remain undocumented because there is detected by the eDarkTrends NER in January 2019. It was subsequently no certainty regarding what quantities are being shipped where and for — identified by NMS labs, but nearly nine month later, in September 2019 what purpose for individual or small network use or for larger scale (NMS Labs, 2019b). This example highlights the utility of the eDark- redistribution. Important future work includes comparing the avail- Trends platform and cryptomarket data in general for timely identifi- ability and shipment origin of substances advertised on the darknet cation and effective surveillance of emerging substance use trends. with the national seizure data to estimate the impact of cryptomarkets Our data suggest that China was the main country of origin of on substance accessibility at the national level. Additional research is wholesale quantities of novel synthetic opioids advertised on needed to monitor novel synthetic opioid sales on the darknet and to DreamMarket. However, 52.6 % of all fentanyl-type drug listings further understand how purchased items are actually being used, either (precursors and pharmaceutical fentanyl excluded) that offered retail- through the analysis of drug forum discussions and/or through online level amounts were posted by unique vendor names who indicated they surveys and interviews with customers and/or sellers (Barratt et al., were shipping from the U.S. and Canada. This highlights the role of 2016). Furthermore, information about new detections of synthetic cryptomarkets in contributing to within-country retail-level novel syn- opioids on cryptomarkets could also be communicated to darknet thetic opioids market transactions (Aldridge and Décary-Hétu, market customers via existing community forums, alongside appro- 2016a,b). Despite the recent crackdown attempts by the Chinese gov- priate harm-reduction advice. Another possibility would be to contact ernment on fentanyl and fentanyl analog production (Associated Press, cryptomarket administrators when information contained in listing 2019; Myers, 2019), the existence of this network of retail sellers lo- descriptions of new substances are erroneous and/or misleading po- cated in the U.S. and Canada calls for an increased surveillance of postal tentially leading to adverse consequences for users. This is particularly packages within and between Canada and the United States (Malaea, true in the case of "synthetic heroin," where buyers are almost com- 2019). In addition, the case of "synthetic heroin" requires further in- pletely unaware of what they are purchasing. vestigation: while representing 44.7 % of the total novel synthetic opioids advertised on DreamMarket, its exact chemical composition Role of funding source remains unknown, potentially increasing the risk for opioid users who ff purchase this type of substance to su er severe adverse consequences. This study was supported by the United States of America National This is of crucial importance considering that "synthetic heroin" ap- Institute on Drug Abuse (NIDA), Grant No. R21 DA-044518-02 peared to be mainly advertised as shipped from Canada (97.9 %). Si- (Daniulaityte, PI; Lamy, PI; Sheth, PI). The funding source had no milar to the U.S., Canada has also experienced notable increases in further role in the study design, in the collection, analysis and inter- novel synthetic opioid availability and associated adverse consequences pretation of the data, in the writing of the report, or in the decision to fi (Pardo et al., 2019). In 2018, novel synthetic opioids were identi ed in submit the paper for publication. more than 50 % of all opioid-related drug samples seized by law enforcement in Canada (Drug Analysis Service, 2019). In addition, 78 % of all overdose deaths that occurred in Canada between January - CRediT authorship contribution statement September 2019, involved fentanyl and/or fentanyl analogs (Special Advisory Committee on the Epidemic of Opioid Overdoses, 2020). Francois R. Lamy: Conceptualization, Data curation, Formal ana- Limitations are noted. First, due to the illegal nature and the vola- lysis, Funding acquisition, Investigation, Methodology, Project admin- tility of cryptomarkets, crawling cryptomarkets represent a unique istration, Software, Supervision, Validation, Writing - original draft, challenge for automated data collection as the research team was un- Writing - review & editing. Raminta Daniulaityte: Conceptualization, able to crawl DreamMarket during some time periods due to Distributed Formal analysis, Funding acquisition, Investigation, Methodology, Denial of Service (DDoS) attacks or because the cryptomarket was Project administration, Resources, Supervision, Validation, Writing - under security maintenance. Second, numbers of vendors were calcu- original draft, Writing - review & editing. Monica J. Barratt: lated based on the sum of vendor "unique names" displayed in each Conceptualization, Funding acquisition, Investigation, Methodology, advertisement, which might not accurately describe the exact number Writing - original draft, Writing - review & editing. Usha Lokala: Data of vendors as one vendor could sell various products using different curation, Methodology, Software, Writing - original draft. Amit Sheth: aliases on one cryptomarket, possibly explaining the large number of Conceptualization, Funding acquisition, Project administration, vendors identified during data collection. Third, this research describes Resources, Supervision, Writing - original draft. Robert G. Carlson: the amounts of novel synthetic opioids available through the listings, Conceptualization, Formal analysis, Funding acquisition, Investigation, and information on the actual amounts of substances sold on Supervision, Validation, Writing - original draft, Writing - review & DreamMarket remain unknown. Forth, shipment origin data were only editing.

7 F.R. Lamy, et al. Drug and Alcohol Dependence 213 (2020) 108115

Declaration of Competing Interest Europol, EMCDDA, 2017. Drugs and the Darknet: Perspectives for Enforcement, Research and Policy. Publications Office of the European Union, Luxembourg. Hall, A.J., Logan, J.E., Toblin, R.L., Kaplan, J.A., Kraner, J.C., Bixler, D., Crosby, A.E., No conflict declared. Paulozzi, L.J., 2008. Patterns of abuse among unintentional pharmaceutical overdose fatalities. JAMA 300, 2613–2620. Hedegaard, H., Warner, M., Miniño, A.M., 2018. Drug Overdose Deaths in the United References States, 1999-2017. US Department of Health and Human Services, Centers for Disease Control and Prevention, NCHS Data Brief, pp. 273. Aldridge, J., Décary-Hétu, D., 2016a. Hidden wholesale: the drug diffusing capacity of Hung, O., Stewart, R.D., Malpas, G., Phipps, S., Hung, D., 2019. Managing the opioid online drug cryptomarkets. Int. J. Drug Policy 35, 7–15. epidemic: back to the basics with resuscitation. CJEM 21, 175–176. Aldridge, J., Décary-Hétu, D., 2016b. Cryptomarkets and the Future of Illicit Drug Jones, C.M., Einstein, E.B., Compton, W.M., 2018. Changes in synthetic opioid involve- Markets. Internet and drug markets, EMCDDA insights, Publications Office of the ment in drug overdose deaths in the United States, 2010-2016. JAMA 319, European Union, Luxembourg, pp. 23–30. 1819–1821. Associated Press, 2019. China to Regulate All Fentanyl Drugs, Keeping Promise to Trump. Kruithof, K., Aldridge, J., Décary-Hétu, D., Sim, M., Dujso, E., Hoorens, S., 2016a. Los Angeles Times. Internet-facilitated Drugs Trade: an Analysis of the Size, Scope and the Role of the Ball, M., Broadhurst, R., Trivedi, H., 2019. How Much Fentanyl Is Available on the Netherlands. R. Corporation. Santa Monica, CA, RAND Corporation. Darknet? Statistical Bulletin 18. Australian Institute of Criminology, Canberra, Kruithof, K., Aldridge, J., Hétu, D.D., Sim, M., Dujso, E., Hoorens, S., 2016b. The Role of pp. 1–6. the “dark Web” in the Trade of Illicit Drugs. RAND Corporation, WODC, Ministerie Barratt, M.J., Aldridge, J., 2016. Everything you always wanted to know about drug van Veilgheid en Justitie. cryptomarkets* (*but were afraid to ask). Int. J. Drug Policy 35, 1–6. Lamy, F.R., Daniulaityte, R., Nahhas, R.W., Barratt, M.J., Smith, A.G., Sheth, A., Martins, Barratt, M.J., Lenton, S., Maddox, A., Allen, M., 2016. ’What if you live on top of a bakery S.S., Boyer, E.W., Carlson, R.G., 2017. Increased in synthetic cannabinoids-related and you like cakes?’-Drug use and harm trajectories before, during and after the harms: results from a longitudinal web-based content analysis. Int. J. Drug Policy 44, emergence of silk road. Int. J. Drug Policy 35, 50–57. 121–129. Barratt, M.J., Aldridge, J., Maddox, A., 2018. Dark Web. The SAGE Encyclopedia of the Lokala, U., Lamy, F.R., Daniulaityte, R., Sheth, A., Nahhas, R.W., Roden, J.I., Yadav, S., Internet, pp. 1–4. Carlson, R.G., 2019. Global trends, local harms: availability of fentanyl-type drugs on Baum, R.J., 2017. Letter to Congress: Response to Questions Concerning Fentanyl. Office the darkweb and accidental overdoses in Ohio. Comput. Math. Organ. Theory 25, of National Drug Control Policy, Washington DC. 48–59. Baumann, M.H., Majumdar, S., Le Rouzic, V., Hunkele, A., Uprety, R., Huang, X.P., Xu, J., Madore, P.H., 2019. Dream Market Shuts Down: Has the DEA Claimed Another Darknet Roth, B.L., Pan, Y.-X., Pasternak, G.W., 2018. Pharmacological characterization of Victim? CCN.com Retrieved 26/03/2019, from. https://www.ccn.com/dream- novel synthetic opioids (NSO) found in the recreational drug marketplace. market-shuts-down-has-the-dea-claimed-another-darknet-victim. Neuropharmacology 134, 101–107. Malaea, M., 2019. Montana Border Patrol Reports Amount of Fentanyl Seized Has Gone Bird, S., Klein, E., Loper, E., 2009. Natural Language Processing with Python. O’Reilly Up By 36 Times Since 2014. Retrieved 20/12/2019, from. Newsweek. https:// Media, Inc. www.ccn.com/dream-market-shuts-down-has-the-dea-claimed-another-darknet- Cameron, D., Smith, G.A., Daniulaityte, R., Sheth, A.P., Dave, D., Chen, L., Anand, G., victim. Carlson, R., Watkins, K.Z., Falck, R., 2013. PREDOSE: a semantic web platform for Mars, S.G., Bourgois, P., Karandinos, G., Montero, F., Ciccarone, D., 2014. “Every ‘never’I drug abuse epidemiology using social media. J. Biomed. Inform. 46, 985–997. ever said came true”: transitions from opioid pills to heroin injecting. Int. J. Drug Carlson, R.G., Nahhas, R.W., Martins, S.S., Daniulaityte, R., 2016. Predictors of transition Policy 25, 257–266. to heroin use among initially non-opioid dependent illicit pharmaceutical opioid Martin, M., Hecker, J., Clark, R., Frye, J., Jehle, D., Lucid, E.J., Harchelroad, F., 1991. users: a natural history study. Drug Alcohol Depend. 160, 127–134. China White epidemic: an eastern United States emergency department experience. Castillo-Carniglia, A., Ponicki, W.R., Gaidus, A., Gruenewald, P.J., Marshall, B.D.L., Fink, Ann. Emerg. Med. 20, 158–164. D.S., Martins, S.S., Rivera-Aguirre, A., Wintemute, G.J., Cerdá, M., 2019. Prescription Mohr, A.L., Friscia, M., Papsun, D., Kacinko, S.L., Buzby, D., Logan, B.K., 2016. Analysis drug monitoring programs and opioid overdoses: exploring sources of heterogeneity. of novel synthetic opioids U-47700, U-50488 and furanyl fentanyl by LC–MS/MS in Epidemiology 30, 212–220. postmortem casework. J. Anal. Toxicol. 40, 709–717. Ciccarone, D., 2019. The triple wave epidemic: supply and demand drivers of the US Musen, M.A., 2015. The Protégé project: a look back and a look forward. AI matters. opioid overdose crisis. Int. J. Drug Policy 71, 183–188. Association of Computing Machinery Specific Interest Group in Artificial Intelligence Ciccarone, D., Ondocsin, J., Mars, S.G., 2017. Heroin uncertainties: exploring users’ 1. https://doi.org/10.1145/2557001.25757003. perceptions of fentanyl-adulterated and-substituted ‘heroin’. Int. J. Drug Policy 46, Myers, S.L., 2019. China Cracks Down on Fentanyl. But Is It Enough to End the U.S. 146–155. Epidemic? The New York Times. Cicero, T.J., Ellis, M.S., Surratt, H.L., Kurtz, S.P., 2014. The changing face of heroin use in NMS Labs, 2019a. Designer Opioids (2018 Scope) (Qualitative). the United States: a retrospective analysis of the past 50 years. JAMA Psychiatry 71, NMS Labs, 2019b. AP-237. Retrieved from:. . https://www.npsdiscovery.org/wp- 821–826. content/uploads/2019/09/AP-237_091619_NMSLabs_Report.pdf. Compton, W.M., Volkow, N.D., 2006. Abuse of prescription drugs and the risk of addic- O’Connor, S., 2017. Fentanyl: China’s deadly export to the United States. US-China tion. Drug Alcohol Depend. 83 (Suppl 1), S4–7. Economic and Security Review Commission, Staff Research Report. Daniulaityte, R., Carlson, R., Brigham, G., Cameron, D., Sheth, A., 2015. "Sub is a weird Pardo, B., Taylor, J., Caulkins, J.P., Kilmer, B., Reuter, P., Stein, B.D., 2019. The Future of drug:" A web-based study of lay attitudes about use of buprenorphine to self-treat Fentanyl and Other Synthetic Opioids. RAND Corporation. opioid withdrawal symptoms. Am. J. Addict. 24, 403–409. Paulozzi, L.J., Budnitz, D.S., Xi, Y., 2006. Increasing deaths from opioid analgesics in the Daniulaityte, R., Juhascik, M.P., Strayer, K.E., Sizemore, I.E., Harshbarger, K.E., United States. Pharmacoepidem. Dr. S. 15, 618–627. Antonides, H.M., Carlson, R.R., 2017a. Overdose deaths related to fentanyl and its Peterson, A.B., Gladden, R.M., Delcher, C., Spies, E., Garcia-Williams, A., Wang, Y., analogs - Ohio, january-february 2017. Morb. Mortal. Rep. 66, 904–908. Halpin, J., Zibbell, J., McCarty, C.L., DeFiore-Hyrmer, J., 2016. Increases in fentanyl- Daniulaityte, R., Lamy, F.R., Smith, G.A., Nahhas, R.W., Carlson, R.G., Thirunarayan, K., related overdose deaths—Florida and Ohio, 2013–2015. Morb. Mortal. Rep. 65, Martins, S.S., Boyer, E.W., Sheth, A., 2017b. “Retweet to pass the blunt”: analyzing 844–849. geographic and content features of cannabis-related tweeting across the United Rosenblum, D., Unick, J., Ciccarone, D., 2020. The rapidly changing US illicit drug States. J. Stud. Alcohol Drugs 78, 910–915. market and the potential for an improved early warning system: evidence from Ohio Daniulaityte, R., Juhascik, M.P., Strayer, K.E., Sizemore, I.E., Zatreh, M., Nahhas, R.W., drug crime labs. Drug Alcohol Depend. 208, 107779. Harshbarger, K.E., Antonides, H.M., Martins, S.S., Carlson, R.R., 2019. Trends in Scholl, L., Seth, P., Kariisa, M., Wilson, N., Baldwin, G., 2019. Drug and opioid-involved fentanyl and fentanyl analogue-related overdose deaths – montgomery County, Ohio, overdose deaths—United States, 2013–2017. Morb. Mortal. Rep. 67, 1419. 2015-2017. Drug Alcohol Depend. 198, 116–120. Special Advisory Committee on the Epidemic of Opioid Overdoses, 2020. Opioid-related DEA, 2018a. Emerging Threat Report, Annual Report 2018. U.S. Drug Enforcement Harms in Canada. Public Health Agency of Canada, Ottawa. https://health-infobase. Administration (DEA). canada.ca/substance-related-harms/opioids. DEA, 2018b. Fentanyl Remains the Most Significant Synthetic Opioid Threat and Poses Unick, G.J., Rosenblum, D., Mars, S., Ciccarone, D., 2013. Intertwined epidemics: national the Greatest Threat to the Opioid User Market in the United States. U.S. Drug demographic trends in hospitalizations for heroin- and opioid-related overdoses, Enforcement Administration (DEA). 1993-2009. PLoS One 8, e54496. DEA, 2019. Emerging Threat Report, First Quarter 2019. U.S. Drug Enforcement Unick, G., Rosenblum, D., Mars, S., Ciccarone, D., 2014. The relationship between us Administration (DEA). heroin market dynamics and heroin-related overdose, 1992-2008. Addiction 109, Dolliver, D.S., 2015. Evaluating drug trafficking on the Tor network: silk Road 2, the 1889–1898. sequel. Int. J. Drug Policy 26, 1113–1123. Van Buskirk, J., Naicker, S., Roxburgh, A., Bruno, R., Burns, L., 2016. Who sells what? Dolliver, D.S., Kuhns, J.B., 2016. The presence of new psychoactive substances in a tor Country specificdifferences in substance availability on the Agora cryptomarket. Int. network marketplace environment. J. Psychoact. Drugs 48, 321–329. J. Drug Policy 35, 16–23. Drug Analysis Service, 2019. Drug Analysis Service: Summary Report of Samples Van Rossum, G., 1990. Python (langage). Analyzed in 2018. Government of Canada. Van Zee, A., 2009. The promotion and marketing of oxycontin: commercial triumph, EMCDDA, 2019. European Drug Report 2019: Trends and Developments. Publications public health tragedy. Am. J. Public Health 99, 221–227. Office of the European Union, Luxembourg.